What is a faster experiment rate worth to your EBITDA?

With agentic software delivery, teams can build, ship and test an idea in days instead of weeks. Roughly one in three tested ideas moves the numbers, so more experiments mean more winners. This calculator uses DORA's published ROI models to put a dollar figure on that for a PE-backed software company.

Added recurring revenue, each year

–

Added EBITDA, each year

–

Engineering capacity freed

–

Over the hold

It stacks up year after year

A winning experiment keeps paying, so each year's winners add to the last. Bars show the EBITDA run-rate at the end of each year.

View as a table

Show the math

How the number is built

Revenue and EBITDA from more experiments

Capacity freed (shown separately, not added to EBITDA)

The formulas come from DORA's ROI of AI-assisted Software Development (2026), which builds on the experiment model in The ROI of DevOps Transformation (2020). The margin step that turns revenue into EBITDA is mine.

Assumptions

Tune it to your business

The defaults come from the DORA research where it has a number, and are marked where it doesn't. Change any of them to match your company.

Want the measured version?

These numbers come from industry benchmarks. I help PE-backed software companies put agentic delivery in place and measure what it actually returns. Copy your numbers and send them over, and I'll help you work out what's real for your teams.

Email Dave

Nothing you enter leaves this page unless you choose to send it.

Sources & method

Where this comes from

DORA and Google Cloud delta, The ROI of AI-assisted Software Development, v2026.1 (2026). Licensed CC BY-NC-SA 4.0. Source of the feature-revenue and capacity formulas and of the defaults for time saved, loaded cost, revenue impact range and J-curve. DORA's own interactive version: dora.dev/ai/roi/calculator.
N. Forsgren, J. Humble, G. Kim, B. Washington, N. Kaul and D. Smith, The ROI of DevOps Transformation, DORA / Google Cloud (2020). Source of the model linking experiment frequency to revenue, and of the performance tiers (Low, Medium, High, Elite) on the experiment sliders.
N. Larsen et al., “Statistical Challenges in Online Controlled Experiments: A Review of A/B Testing Methodology” (2023). Cited by both reports for the finding that about one third of well-designed features improve key metrics.

What this counts, and what it doesn't

It's an estimate to start a conversation, not a forecast. DORA describes its own calculator as “a high-uncertainty estimate.”
The revenue impact per winning experiment drives most of the result. DORA suggests 0.01% to 1%. I default to 0.1% to stay conservative.
It doesn't model diminishing returns. At very high experiment rates, ideas may get smaller, so the gain is likely overstated there.
Capacity is time, not cash. It only becomes EBITDA if you avoid hiring, so it's never added to the EBITDA figure. Like DORA, I recommend reinvesting it rather than cutting headcount.
The 80% incremental margin is my assumption for SaaS revenue, not a DORA figure.
It doesn't net out costs. AI tools, training, the productivity dip while teams adapt, and any change in stability or downtime aren't included. DORA's calculator covers those.

An independent page by Dave at Mangoteque. Not affiliated with or endorsed by Google, Google Cloud or DORA. The formulas and benchmark figures are used with attribution; any errors in applying them are mine.